PhD course by Marco Fanfani

Abstract

Being able to extract accurate measurements of a three-dimensional scene is a fundamental task in several applications spanning from autonomous vehicles to industrial applications. For this reason, Geometrical Computer Vision is an active research field: on the one hand, the widely used classical model-based algorithms can offer accurate and controllable solutions; on the other hand, the more recently emerged deep-learning based approaches try to offer alternatives to deal with 3D applications.
In this course, Computer Vision techniques working in three-dimensional environments will be presented and discussed. Initially, the more relevant concepts related to projective geometry, camera models, and multi-view computer vision – essential to comprehend the 3D related applications that will be discussed in this course – will be introduced. Then, 3D reconstruction algorithms will be presented and an overview on visual odometry and on simultaneous localization and mapping will be given, including both model-based and learning-based approaches. During the course, the theoretical presentations will be sided with examples of ready-to-use practical solutions.